jjfoooo4

(I work for Block, not on this)

Cash App was not an acquisition, it originated as a hackweek within (what was then called) Square.

Possibly you’re thinking of the Afterpay acquisition?

A sticky product is one that switching away from creates a major hassle. Which means the user will pay more to avoid said hassle.

“I don’t really have a strong preference between the two” is another way of saying “the product isn’t sticky”, which is another way of saying “this provider has very little room to increase margins”

I don’t really get the business plan for open weights model companies, is the idea companies would pay them for serving?

One major difference is that to send network traffic, you have to go through an ISP. With AI you have the option of local open source models

Down from 10 years ago, down quite a bit from 20 years ago

But there's not an everyday use case for firearms and explosives for the average person. There's civilian, everyday task that would be easier were better firearms and explosives available

Surely both manufacturer and convenience store have incentive to push the addictive product in this scenario.

In any case, drug dealers really don’t need to do any pushing, the drugs sell themselves. Have you ever taken an opioid? The idea that unfettered access would result in less addiction and death is a pretty remarkable POV

That they aren’t participating in the AI arms race is very good for current earning

Of course, this then introduces the circular reasoning "because of a potential US attack", but of course if Iran wasn't funding Hamas, Hezbollah, Houthis

I think the first step of thinking about war objectively is to consider how each side sees it. The US POV is no less circular, from Iran’s perspective - they could list any number of provocations from the US to justify arming themselves, none more obvious than the war itself.

The debate around who started the hostility is ultimately pointless, the question is what to do about. Ideally the answer isn’t “arm for obliteration because the other side started it”

Even as someone highly interested in memory I don’t see it as a useful tool for coding. The source of truth for what a repo does or should do is the repo itself.

What you’re describing sounds more like code review guidelines, which can be explicitly brought into context at specific times during a change. A memory system is both too complex and less accurate for this

Simple vector similarity plus a cheap model to filter results works pretty well. Though ofc t does add tokens to your primary chat, which is the basic tradeoff of memory systems in general (in addition to latency)

I favor automatic recall, invisible to the agent. For memory creation, I find tool calls do a pretty good job, though I also like automatic memory creation on context compression.

I think with automatic creation you need async consolidation (calling it dreaming is a little dramatic for my taste).

My implementation is at Elroy.bot, I recently wrote about different approaches to agent memory here: https://tombedor.dev/approaches-to-agent-memory/

But these products are all drop in replacements for each other. I've recently favored Codex more than CC, just because rate limits got mildly annoying. I really didn't have to change anything about my workflow in doing that.

My feeling is that if the AI is this good, the audience will just prompt the AI themselves and cut out the middleman.

I call this the instant imitator trap. If anything AI generates stands out from the slop, the slop generators will just imitate it, thus quickly making whatever standout quality from the "original" work also slop.

I wrote about it here: https://tombedor.dev/creativity/

From the press release, it's not all that clear what Deezer is doing about it. 44% of uploads getting less than 1% of non-fraudulent streams seems like a pretty strong reason to outright ban AI generated submissions.

For the non-fraudulent listens, I'm very curious how many of these are part of auto-generated playlists. Are people just being served this music as part of a feed, or are they actually seeking it out? I'd be very surprised if it was the latter.

It might be a bias in terms of the probability of events, but I'm not so sure this is a market inefficiency in terms of actual trading strategy. If true odds are 1% and the event is priced at 4%, I can sell NO for a 3% edge... but lose 100% once out of a hundred. Doesn't seem worth it!

Which I'm confused about - wouldn't decreasing the cache TTL increase compute demand?

AI is a floor raiser, not a ceiling raiser

you're going to end up systematically misinformed on political topics that people find outrageous.

That sounds... fine?

I vibe coded a charting library for excalidraw that I’m working into my blog:

https://github.com/tombedor/excalicharts

VC can spend all the money in the world and it won't matter if the cost of switching providers is effectively zero.

If I want to switch from Windows to Linux, I have to reconsider a whole variety of applications, learn a different UX, migrate data, all sorts of annoyances.

When I switch between Codex and Claude Code, there is literally no difference in how I interact with them. They and a number of other competitors are drop in replacements for each other.

That is 100% what I think is going to happen.

I wrote about it here: https://tombedor.dev/open-source-models/

people might luck into something genuinely good and worth sharing from time to time.

A) it would be impossible to find in a sea of AI generated slop

B) even if it were to be recognized as good, it would be instantly copied by other AI’s such that it would be very shortly thereafter be also considered slop

For any work to gain traction with an audience, there needs to be scarcity. Art and artists are valued because they are unique in some way, something about it or them cannot be replicated by others. The ability to instantly produce a piece of “art” negates any artistic value, at least as far as audiences are concerned.

Claude code is starting to develop one as people start to build things that require Claude Code specifically, not just any LLM.

I hate to be a "source?" guy, but I'm curious if you have any examples of this. Skills and MCP are really the only extensions on CC itself I'm aware of, and these are both supported in Codex.

Things like Dispatch / remote sessions is something CC has that Codex does not, but these features are quite easy to replicate (and I expect Codex to do so in short order).

Nothing about AI mindshare is durable. As Anthropic tightens rate limits and/or raises prices, everyone will move to something else.

This week I’ve been hitting CC rate limits, and I switched to Codex with virtually no disruption to my workflow. It’s not good for Anthropic that I was able to do that!

Both companies have the same issue: the unit economics are bad, and there’s no moat.

I’ve been primarily on Claude for the last 6 months or so, but have been hitting rate limits. I switched work to Codex seamlessly, just like I could switch to any other provider seamlessly

Curious if you agree that local is where Apple's bet is long term - it's out of reach now, but I found the jump in capability for the top line laptop interesting. Presumably chip development hasn't focused all that much on running LLM's for all that long, I'm wondering what kind of jump we'll see two or three releases down the line.

Would be interested in collaborating! My email is in my bio

I’m really, really uninterested in reading AI content that other people have generated. If I’m on Twitter, I’m looking for what humans have to say.

I've been playing with something similar! Curious how you are interfacing with Chess.com / lichess.org, are you having people paste their PGN's?